Re: [PATCH V2 1/5] X86/Hyper-V: Add flush HvFlushGuestPhysicalAddressSpace hypercall support

[Tianyu Lan]

Hi Michael:
	Thanks for your review.

On 7/11/2018 5:29 AM, Michael Kelley (EOSG) wrote:
> From: Tianyu Lan <Tianyu.Lan@xxxxxxxxxxxxx> Monday, July 9, 2018 2:03 AM
>> Hyper-V supports a pv hypercall HvFlushGuestPhysicalAddressSpace to
>> flush nested VM address space mapping in l1 hypervisor and it's to
>> reduce overhead of flushing ept tlb among vcpus. This patch is to
>> implement it.
>>
>> Signed-off-by: Lan Tianyu <Tianyu.Lan@xxxxxxxxxxxxx>
>> ---
>>   arch/x86/hyperv/Makefile           |  2 +-
>>   arch/x86/hyperv/nested.c           | 64 ++++++++++++++++++++++++++++++++++++++
>>   arch/x86/include/asm/hyperv-tlfs.h |  8 +++++
>>   arch/x86/include/asm/mshyperv.h    |  2 ++
>>   4 files changed, 75 insertions(+), 1 deletion(-)
>>   create mode 100644 arch/x86/hyperv/nested.c
>> +#include <linux/types.h>
>> +#include <asm/hyperv-tlfs.h>
>> +#include <asm/mshyperv.h>
>> +#include <asm/tlbflush.h>
>> +
>> +int hyperv_flush_guest_mapping(u64 as)
>> +{
>> +	struct hv_guest_mapping_flush **flush_pcpu;
>> +	struct hv_guest_mapping_flush *flush;
>> +	u64 status;
>> +	unsigned long flags;
>> +	int ret = -EFAULT;
>> +
>> +	if (!hv_hypercall_pg)
>> +		goto fault;
>> +
>> +	local_irq_save(flags);
>> +
>> +	flush_pcpu = (struct hv_guest_mapping_flush **)
>> +		this_cpu_ptr(hyperv_pcpu_input_arg);
>> +
>> +	flush = *flush_pcpu;
>> +
>> +	if (unlikely(!flush)) {
>> +		local_irq_restore(flags);
>> +		goto fault;
>> +	}
>> +
>> +	flush->address_space = as;
>> +	flush->flags = 0;
>> +
>> +	status = hv_do_hypercall(HVCALL_FLUSH_GUEST_PHYSICAL_ADDRESS_SPACE,
>> +				 flush, NULL);
> 
> Did you consider using a "fast" hypercall?  Unless there's some reason I'm
> not aware of, a "fast" hypercall would be perfect here as there are 16 bytes
> of input and no output. Vitaly recently added hv_do_fast_hypercall16()
> in the linux-next tree. See __send_ipi_mask() in hv_apic.c in linux-next
> for an example of usage.  With a fast hypercall, you don't need the code for
> getting the per-cpu input arg or the code for local irq save/restore, so the
> code that is left is a lot faster and simpler.
> 
> Michael
> 

Good suggestion. But the "fast" hypercall still is not available in 
kvm-next branch and it's in the x86 tip repo. We may rework this with 
"fast" hypercall in the next kernel development cycle if this patchset 
is accepted in for 4.19.

>> +	local_irq_restore(flags);
>> +
>> +	if (!(status & HV_HYPERCALL_RESULT_MASK))
>> +		ret = 0;
>> +
>> +fault:
>> +	return ret;
>> +}
>> +EXPORT_SYMBOL_GPL(hyperv_flush_guest_mapping);